EyeMVP: OCT-Informed Fundus Representation Learning via Paired CFP--OCT Pretraining
Researchers have developed EyeMVP, a novel foundation model for retinal image analysis that integrates data from both color fundus photography (CFP) and optical coherence tomography (OCT). Pretrained on a large dataset of paired CFP-OCT images from over 112,000 patients, EyeMVP learns to enhance CFP representations with OCT-derived structural information. This allows for more accurate diagnoses using only CFP images during inference, showing improved performance on tasks like macular edema and myopic macular schisis detection compared to existing models and human ophthalmologists in exploratory studies. AI